Add VoxConverse recipe (#1142)

* Add VoxConverse recipe

* Do not resplit the dataset by default

---------

Co-authored-by: Piotr Żelasko <[email protected]>

docs/corpus.rst

@@ -177,6 +177,8 @@ a CLI tool that create the manifests given a corpus directory.
     - :func:`lhotse.recipes.prepare_vctk`
   * - VoxCeleb
     - :func:`lhotse.recipes.prepare_voxceleb`
+  * - VoxConverse
+    - :func:`lhotse.recipes.prepare_voxconverse`
   * - VoxPopuli
     - :func:`lhotse.recipes.prepare_voxpopuli`
   * - WenetSpeech

lhotse/bin/modes/recipes/__init__.py

@@ -72,6 +72,7 @@
 from .uwb_atcc import *
 from .vctk import *
 from .voxceleb import *
+from .voxconverse import *
 from .voxpopuli import *
 from .wenet_speech import *
 from .xbmu_amdo31 import *

lhotse/bin/modes/recipes/voxconverse.py

@@ -0,0 +1,27 @@
+import click
+
+from lhotse.bin.modes import download, prepare
+from lhotse.recipes.voxconverse import download_voxconverse, prepare_voxconverse
+from lhotse.utils import Pathlike
+
+
+@download.command(context_settings=dict(show_default=True))
+@click.argument("target_dir", type=click.Path())
+@click.option("--force-download", is_flag=True, default=False, help="Force download")
+def voxconverse(target_dir: Pathlike, force_download=False):
+    """VoxConverse dataset download."""
+    download_voxconverse(target_dir, force_download=force_download)
+
+
+@prepare.command(context_settings=dict(show_default=True))
+@click.argument("corpus_dir", type=click.Path(exists=True, dir_okay=True))
+@click.argument("output_dir", type=click.Path())
+@click.option(
+    "--split-test",
+    is_flag=True,
+    default=False,
+    help="Split test part into dev and test parts",
+)
+def voxconverse(corpus_dir: Pathlike, output_dir: Pathlike, split_test: bool = False):
+    """VoxConverse data preparation."""
+    prepare_voxconverse(corpus_dir, output_dir=output_dir, split_test=split_test)

lhotse/recipes/__init__.py

@@ -71,6 +71,7 @@
 from .uwb_atcc import download_uwb_atcc, prepare_uwb_atcc
 from .vctk import download_vctk, prepare_vctk
 from .voxceleb import download_voxceleb1, download_voxceleb2, prepare_voxceleb
+from .voxconverse import download_voxconverse, prepare_voxconverse
 from .voxpopuli import download_voxpopuli, prepare_voxpopuli
 from .wenet_speech import prepare_wenet_speech
 from .xbmu_amdo31 import download_xbmu_amdo31, prepare_xbmu_amdo31

lhotse/recipes/voxconverse.py

@@ -0,0 +1,149 @@
+"""
+VoxConverse is an audio-visual diarisation dataset consisting of multispeaker clips of human speech, extracted from YouTube videos.
+Updates and additional information about the dataset can be found at our website (https://www.robots.ox.ac.uk/~vgg/data/voxconverse/index.html).
+"""
+
+import json
+import logging
+import re
+import shutil
+import zipfile
+from collections import defaultdict
+from concurrent.futures import ProcessPoolExecutor
+from pathlib import Path
+from typing import Dict, Iterable, Optional, Union
+from urllib.error import HTTPError
+
+from tqdm.auto import tqdm
+
+from lhotse import fix_manifests, validate_recordings_and_supervisions
+from lhotse.audio import Recording, RecordingSet
+from lhotse.supervision import AlignmentItem, SupervisionSegment, SupervisionSet
+from lhotse.utils import Pathlike, is_module_available, resumable_download
+
+DEV_AUDIO_ZIP = (
+    "https://www.robots.ox.ac.uk/~vgg/data/voxconverse/data/voxconverse_dev_wav.zip"
+)
+TEST_AUDIO_ZIP = (
+    "https://www.robots.ox.ac.uk/~vgg/data/voxconverse/data/voxconverse_test_wav.zip"
+)
+ANNOTATIONS_ZIP = "https://github.com/joonson/voxconverse/archive/master.zip"
+
+
+def download_voxconverse(
+    corpus_dir: Pathlike,
+    force_download: bool = False,
+):
+    corpus_dir = Path(corpus_dir)
+    corpus_dir.mkdir(parents=True, exist_ok=True)
+    completed_detector = corpus_dir / ".completed"
+
+    if not completed_detector.is_file() or force_download:
+        print("Downloading VoxConverse dev set")
+        resumable_download(DEV_AUDIO_ZIP, corpus_dir / "dev.zip")
+        with zipfile.ZipFile(corpus_dir / "dev.zip") as zip_f:
+            zip_f.extractall(corpus_dir / "dev")
+
+        shutil.copytree(
+            corpus_dir / "dev/audio", corpus_dir / "dev", dirs_exist_ok=True
+        )
+        shutil.rmtree(corpus_dir / "dev/audio")
+
+        print("Downloading VoxConverse test set")
+        resumable_download(TEST_AUDIO_ZIP, corpus_dir / "test.zip")
+        with zipfile.ZipFile(corpus_dir / "test.zip") as zip_f:
+            zip_f.extractall(corpus_dir / "test")
+
+        shutil.copytree(
+            corpus_dir / "test/voxconverse_test_wav",
+            corpus_dir / "test",
+            dirs_exist_ok=True,
+        )
+        shutil.rmtree(corpus_dir / "test/voxconverse_test_wav")
+
+        print("Downloading VoxConverse annotations")
+        resumable_download(ANNOTATIONS_ZIP, corpus_dir / "annotations.zip")
+        with zipfile.ZipFile(corpus_dir / "annotations.zip") as zip_f:
+            zip_f.extractall(corpus_dir)
+
+        shutil.copytree(
+            corpus_dir / "voxconverse-master", corpus_dir, dirs_exist_ok=True
+        )
+        shutil.rmtree(corpus_dir / "voxconverse-master")
+
+        # cleanup
+        (corpus_dir / "dev.zip").unlink()
+        (corpus_dir / "test.zip").unlink()
+        (corpus_dir / "annotations.zip").unlink()
+        completed_detector.touch()
+
+    print("Done")
+
+
+def prepare_voxconverse(
+    corpus_dir: Pathlike,
+    output_dir: Optional[Pathlike] = None,
+    split_test: bool = False,  # test part is larger than dev part - split it into dev and test by default
+):
+    corpus_dir = Path(corpus_dir).absolute()
+
+    splits = {}
+    if split_test:
+        splits["train"] = sorted((corpus_dir / "dev").glob("*.wav"))
+        test_files = sorted((corpus_dir / "test").glob("*.wav"))
+        splits["dev"] = test_files[: len(test_files) // 2]
+        splits["test"] = test_files[len(test_files) // 2 :]
+    else:
+        splits["dev"] = sorted((corpus_dir / "dev").glob("*.wav"))
+        splits["test"] = sorted((corpus_dir / "test").glob("*.wav"))
+
+    manifests = {}
+    for subset, wavs in splits.items():
+        recordings = []
+        supervisions = []
+        for wav_file in wavs:
+            recordings.append(Recording.from_file(wav_file))
+            rttm_file = wav_file.with_suffix("").with_suffix(".rttm")
+            for ix, (start, duration, speaker) in enumerate(_read_rttm(rttm_file)):
+                supervisions.append(
+                    SupervisionSegment(
+                        id=f"{wav_file.stem}-{ix}",
+                        recording_id=wav_file.stem,
+                        start=start,
+                        duration=duration,
+                        channel=0,
+                        language="en",
+                        speaker=speaker,
+                    )
+                )
+
+        recording_set = RecordingSet.from_recordings(recordings)
+        supervision_set = SupervisionSet.from_segments(supervisions)
+        recording_set, supervision_set = fix_manifests(recording_set, supervision_set)
+        validate_recordings_and_supervisions(recording_set, supervision_set)
+
+        if output_dir is not None:
+            output_dir = Path(output_dir)
+            output_dir.mkdir(parents=True, exist_ok=True)
+            recording_set.to_file(
+                output_dir / f"voxconverse_recordings_{subset}.jsonl.gz"
+            )
+            supervision_set.to_file(
+                output_dir / f"voxconverse_supervisions_{subset}.jsonl.gz"
+            )
+
+        manifests[subset] = {
+            "recordings": recording_set,
+            "supervisions": supervision_set,
+        }
+
+    return manifests
+
+
+def _read_rttm(filename):
+    with open(filename, "r") as f:
+        for line in f.readlines():
+            line = line.strip()
+            if line.startswith("SPEAKER"):
+                _, _, _, start, duration, _, _, speaker, _, _ = line.split()
+                yield float(start), float(duration), speaker